User Warning Transformer with Torch

[hdaipteam]

How to reproduce the behaviour

nlp = spacy.load("en_core_web_sm")

Your Environment

• Operating System: Ubuntu 22.04
• Python Version Used: 3.11
• spaCy Version Used: 3.7.2 - latest / 4.0.0-dev
• Environment Information: IPyNotebook

I cought the warning if I am loading the pre-defined model in both version of SpaCy Latest / Dev

python3.11/site-packages/transformers/utils/generic.py:441: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.
_torch_pytree._register_pytree_node(

[jefhil]

I assume this is harmless but users see warnings like this and get worried. Any idea when this warning will go away?

[nablabits]

It seems that this was fixed back on Jan 3rd (source). It appears to me that Spacy does not require transformers (source), however spacy-transformers does (source) and allows for transformers >=3.4.0,<4.42.0 . The commit that fixed this belongs to 4.37.0 so @hdaipteam you might want to bump your local version of transformers

[Bonnevie]

@nablabits I have been trying to fix the warning this morning, but it seems that the current latest version of spacy-transformers doesn't have the relaxed constraint that is in the master branch, but rather has <4.37.0? I got this commit from the version tag: https://github.com/explosion/spacy-transformers/blob/ba6bf8202662f59aa1862bd4c5dd6fd0119a3335/requirements.txt

[nablabits]

Hey @Bonnevie, sorry the late reply, it appears to me that you are pointing to the commit on Apr 23rd, whereas mine points to May 30th, source, which is the latest one. Does this make sense on your end?